5  Results

In this section, study population characteristics and findings from analyses will be presented.

5.1 Study I

5.1.1 Descriptive

In The Maastricht Study, 7,449 participants were included between November 2010 and December 2020, of whom 1,316 reported a history of CVD.1 A total of 4,379 participants had valid 24-hour HRV measurements, and among these, 3,673 had a valid measurement of either CD or cf-PWV. Study population included 3673 participants. The study population represented diabetes risk of NGM (65%), prediabetes (15%), and T2D (20%). The median (IQR) cf-PWV (aortic stiffness) became higher with diabetes status: NGM: 8.08 m/s (7.28, 9.16), prediabetes: 8.96 m/s (7.84, 10.32), and T2D: 9.36 m/s (8.16, 10.80). CD (carotid stiffness) decreased: NGM: 15.0 (11.8, 18.8), prediabetes: 13.5 (10.4, 16.9), and T2D: 12.5 (9.9, 16.0) × 10⁻³/kPa. SDNN (ms) was highest in NGM and lowered with worsening glucose metabolism: NGM: 138ms (117, 164), prediabetes: 127ms (106, 152), and T2D: 116ms (96, 139). Further description of characteristics by diabetes are described in Table 5.1.

Table 5.1: Study charteristics in The Maastricht Study (Study I) by diabetes status
Characteristic Normal glucose metabolism N = 2,389 Prediabetes N = 538 Type 2 Diabetes N = 746
Women 1,361 (57%) 258 (48%) 265 (36%)
Age (years) 58 (51, 64) 62 (57, 68) 63 (57, 68)
Moderate to vigorous physical activity (hours/week) 5.3 (3.0, 8.3) 4.5 (2.3, 7.5) 3.8 (1.5, 6.8)
BMI (kg/m²) 25.0 (22.9, 27.4) 27.2 (24.9, 30.1) 28.8 (26.0, 31.7)
HbA1c (%) 5.35 (5.17, 5.63) 5.63 (5.35, 5.90) 6.54 (6.08, 7.09)
HDL (mmol/L) 1.60 (1.30, 2.00) 1.40 (1.20, 1.80) 1.30 (1.00, 1.50)
Total cholesterol (mmol/L) 5.50 (4.80, 6.20) 5.50 (4.80, 6.30) 4.50 (3.90, 5.20)
Triglycerides (mmol/L) 1.05 (0.80, 1.45) 1.39 (1.03, 1.90) 1.51 (1.08, 2.14)
Duration of type-2 diabetes (only for diagnosed participants) 3 (0, 9)
Mean IBI (ms) 838 (775, 907) 815 (760, 897) 806 (744, 889)
SDNN (ms) 138 (117, 164) 127 (106, 152) 116 (96, 139)
RMSSD (ms) 26 (21, 34) 24 (19, 33) 22 (17, 31)
TP (ms²) 12,596 (8,880, 17,498) 10,615 (7,134, 15,374) 8,880 (6,064, 12,722)
ULF (ms²) 10,771 (7,392, 15,142) 8,948 (5,852, 13,374) 7,524 (5,036, 11,001)
VLF (ms²) 1,198 (833, 1,692) 1,015 (685, 1,478) 816 (541, 1,267)
HF (ms²) 94 (57, 158) 78 (47, 138) 63 (36, 117)
Systolic blood pressure (mmHg) 123 (114, 133) 129 (122, 140) 130 (122, 139)
Diastolic blood pressure (mmHg) 75 (71, 80) 78 (73, 83) 76 (72, 81)
Mean arterial pressure (mmHg) 95 (88, 102) 99 (93, 107) 98 (92, 105)
Carotid artery distensibility (10-3/kPa) 15.0 (11.8, 18.8) 13.5 (10.4, 16.9) 12.5 (9.9, 16.0)
Carotid-femoral pulse wave velocity (m/s) 8.08 (7.28, 9.16) 8.96 (7.84, 10.32) 9.36 (8.16, 10.80)
Glucose-lowering medication 0 (0%) 0 (0%) 519 (70%)
Antihypertensive medication 431 (18%) 199 (37%) 478 (64%)
Beta blockers 149 (6.2%) 77 (14%) 195 (26%)
Lipid-lowering medication 280 (12%) 141 (26%) 484 (65%)
Note:
Categorical variables: N (%); Continous variables: Median (IQT:25th-75th). Table are adapted from supplementary material in Appendix Study I.

5.1.2 24-hour HRV and arterial stiffness

Time-domain HRV

In the fully adjusted model 2, cf-PWV was 2.8% (CI: 2.1; 3.4) lower, while CD was 3.3% (CI: 1.5; 5.1) higher per SD higher in HRV time-domain Z-score (see see Figure 5.1 A and B). Among the time-domain indices, SDNN, SDNNi, and SDANN showed the strongest associations, with cf-PWV being lower by 2.5% (CI: 2.0; 3.1), 2.5% (CI: 1.9; 3.4), and 2.2% (CI: 1.7; 2.7), respectively.1 Conversely, CD was higher by 3.2% (CI: 1.7; 4.7), 3.0 % (CI: 1.4; 4.6), and 2.8% (CI: 1.3; 4.3), respectively. RMSSD and pNN50 showed a weaker association with cf-PWV (-1.1% [CI: -1.4; -0.4], and -1.1 [-1.7; -0.6]), while no evidence for an association was found with CD.1

Frequency-domain HRV

In the fully adjusted model 2, cf-PWV was 2.8% (CI: 2.1; 3.5) lower, while CD was 3.2% (CI: 1.3; 5.1) higher per SD higher in HRV frequency-domain Z-score (see see Figure 5.1 C and D). Among the frequency-domain indices, total power, VLF, and ULF showed the strongest associations, with cf-PWV being lower by 2.2% (CI: 1.7; 2.8 ), 2.4% (CI: 1.9; 4.0), and 2.1% (CI: 1.5; 2.6), respectively.1 Conversely, CD was higher by 2.7% (CI: 1.2; 4.2), 2.4% (CI: 0.9; 4.1), and 2.6% (CI: 1.1; 4.1), respectively. HF showed a weaker association with cf-PWV (-0.9% [CI: -1.4; -0.4]), while no evidence for an association was found with CD. Mean interbeat interval was associated with 2.4 % (CI: 1.8; 2.9) lower cf-PWV and 4.5% (3.1; 6.1) higher CD.1

A: Percentage PWV per SD in time-domain composite z-score B: Percentage PWV per SD in frequency-domain composite z-score C: Percentage higher CD per SD in time-domain composite z-score D: Percentage CD per SD in frequency-domain composite z-score. All regression lines were adjusted for being a male, 60 years old, low educational level, without prediabetes or type-2 diabetes, and with 96mmHg MAP. From Figure 1 in Appendix Study I1

Figure 5.1: Association between HRV and arterial stiffness

5.1.3 Effect modification of diabetes status

The association between HRV time-domain Z-scores and cf-PWV and CD was significantly modified by prediabetes (cf-PWV: -4.9 [CI: -6.5; -3.2] \(^{interaction(*) ^{p-value< 0.01}}\) CD: 8.0 [CI:3.8; 12.5]\(^{*^{p-value< 0.01}}\)) but not by T2D (cf-PWV: -3.5 % [CI: -4.8; -2.1)] \(^{*^{p-value< 0.1}}\) CD: 4.8 % [CI:1.3; 8.4]\(^{*^{p-value< 0.1}}\))1. For the indices SDNN and SDANN, the association with both cf-PWV and CD was significantly modified by both prediabetes and T2D1.

The association between HRV frequency-domain Z-score and cf-PWV was statistically significant modified from NGM by prediabetes (-5.7 %[CI:-7.4; -3.9]\(^{*^{p-value< 0.01}}\)) and T2D (-3.9 %[CI:-5.4; -2.3]\(^{*^{p-value< 0.05}}\)) while CD was only modified by prediabetes (8.3 %[CI:3.6; 13.2]\(^{*^{p-value< 0.01}}\)) but not by T2D (5.3 %[CI:1.4; 9.4]\(^{*^{p-value< 0.1}}\))1. For the indices total power and ULF, the association with both cf-PWV and CD was statistically significant modified by both prediabetes and T2D. Mean inter beat interval association with cf-PWV or CD was not significantly modified by diabetes status1.

As no stepwise increase was observed in the modification of glucose metabolism status from prediabetes to T2D,the subgroup with T2D was excluded to test whether the association was gradually modified by dysglycemia. In this subgroup, the association between HRV time and frequency-domain Z-scores and measures of arterial stiffness was modified by HbA1c (range of interaction p-values: 0.1 to 0.005) (see Figure Figure 5.2). For example, per SD lower HRV frequency domain Z-score at HbA1c 6.4% was associated with a 5.4% higher (CI: 3.5; 7.2) cf-PWV, which was 2.0% to 4.0% higher compared to the magnitude of association at HbA1c levels of 5.6% and 4.8% (see see Figure 5.2 B). In CD, per SD lower HRV frequency domain Z-score at HbA1c 6.4% was associated with an 8.1% lower (CI: -13.5; -2.9) CD, which was 4.8% to 9.5% lower compared to the magnitude of association at HbA1c levels of 5.6% and 4.8% (see Figure 5.2 D). No association between HRV frequency domain Z-score and CD was observed at HbA1c levels between 4.8% and 5.2%.

A: Percentage PWV per SD in time-domain composite z-score B: Percentage PWV per SD in frequency-domain composite z-score C: Percentage higher CD per SD in time-domain composite z-score D: Percentage CD per SD in frequency-domain composite z-score. Model adjusted for sex, age, educational status, diabetes status, and MAP, physical activity, smoking behaviour, alcohol use, body mass index, hba1c, triglycerides, total-to-high density lipoprotein cholesterol ratio, lipid-modifying- and antihypertensive medication. Figure was based on data in Study I1

Figure 5.2: Association between HRV and arterial stiffness modified by HbA1c in subpopulation without T2D

5.2 Study II

5.2.1 Descriptive

The ADDITION-PRO population consisted of 1,627 participants with at least 48-hour HRV measures, while 1,432 had all hours represented with hourly HRV and physical acceleration. The study population included different tiers of diabetes risk: 154 individuals at low risk (9%), 889 at high risk (51%), 314 with impaired fasting glucose (IFG) (18%), 226 with impaired glucose tolerance (IGT) (13%), and 161 with both IFG and IGT (9%). SDNN was categorized into five groups: very-low (SDNN< 100 ms), low (SDNN 100-120 ms), middle (SDNN 121-140 ms), high (SDNN 141-160 ms) and very-high (SDNN >160 ms).

Characteristics are described in ?tbl-add. Participants in the lowest SDNN group (<100 ms) were older (Mean ± SD) (67.4 ± 6.9 years), had higher BMI (28.1 ± 5.4), HbA1c (5.9 ± 0.9), triglycerides (1.5 ± 0.9 mmol/L), and rHR (67.8 ± 5.7 bpm), were more likely to use anti-hypertensive medication (61%), and had lower physical activity energy expenditure (46.8 ± 24.0 kJ/day) compared to those with higher SDNN levels.

?(caption)

5.2.2 Multiday HRV and MACE, heart failure, and all-cause mortality.

The mean (SD) multiday SDNN was 139.0 (32.3) ms, and the mean heart rate was 73.5 (9.1) bpm.2 In the fully adjusted model, SDNN per SD was associated with a lower incidence rate ratio (IRR) for MACE 0.82 (CI: 0.69; 0.97), heart failure 0.76 (CI: 0.58; 0.99), and mortality rate ratio of 0.79 (CI: 0.66; 0.94).2 In model with pre-adjustment for rHR, the proportion of the association explained between HRV and MACE, HF, and all-cause mortality was 14%, 25%, and 19%, respectively.2 When knots were included in the model, the risk was found to be higher as SDNN dropped below approximately 120–110 ms (around the 20th percentile), suggesting a potential threshold for elevated risk.2 Therefore, age-specific IR were calculated at SDNN levels of 100 ms, 120 ms, and 160 ms, respectively.

Multiday SDNN levels (100 ms, 120 ms, 160 ms) by age-specific incidence rates for A) major adverse cardiovascular events, B) heart failure hospitalisation, and C) all-cause mortality. Model adjusted for age, sex, education, alcohol consumption, smoking behavior, physical activity, body mass index, total cholesterol, and Hba1c. Figure was based on data in Study II2

Figure 5.3: Multiday SDNN and mean HR association with major adverse cardiovascular events, heart failure, and all-cause mortality

At age 65, the IR per 1000 person-years for MACE was 44.2 (CI: 33.5; 58.3) at SDNN = 100 ms, which was higher than the rates observed at SDNN = 120 ms (IR: 37.6 [CI: 29.2; 48.3]) and SDNN = 160 ms (IR: 34.7 [CI: 27.0; 44.5]) (Figure 5.3 A). The IR was observed to become higher with age, reaching its peak at age 72. For heart failure at age 65, the IR was 20.5 (CI: 15.3; 27.5) at SDNN = 100 ms, slightly higher than at SDNN = 120 ms (IR: 19.6 [CI: 14.8; 25.9]) and SDNN = 160 ms (IR: 19.8 [CI: 15.0; 26.2]) (Figure 5.3 B). The IR remained stable until age 70, after which it became higher. For all-cause mortality at age 65, the IR was 11.6 (CI: 6.3; 21.4) at SDNN = 100 ms, higher than at SDNN = 120 ms (IR: 7.6 [CI: 4.3; 13.3]) and SDNN = 160 ms (IR: 6.0 [CI: 3.4; 10.4]) (Figure 5.3 C).

5.2.3 Hourly HRV and MACE, heart failure, and all-cause mortality.

From the hourly recordings, a clear periodicity in SDNN, heart rate, sleep patterns, and physical acceleration was observed (see Figure S4 in supplemental material Appendix Study II). Mean (SD) SDNN was found to increase from 5–6 AM (70.2 [28.8] ms), peaking at 8–9 AM (92.1 [29.0] ms), followed by a gradual decline, reaching its lowest point around 2 AM the next day (64.1 [28.1] ms).2 A similar circadian pattern was observed in heart rate, although its peak was reached two hours later, starting at 9 AM (76.7 [10.9] bpm).2 After peaking, heart rate was observed to remain stable throughout the afternoon before gradually decreasing.2

In Figure 5.4, hourly SDNN (preadjusted for heart rate and physical acceleration), heart rate, and physical acceleration associations were examined. Models were adjusted for age, sex, education, alcohol consumption, smoking behavior, BMI, total cholesterol, and HbA1c. The morning response of SDNN was found to be most indicative of MACE, with the strongest association observed from 6–7 AM (IRR: 0.84; 95% CI: 0.71–1.00 per SD higher SDNN) (see Figure 5.4 A).2 Heart rate between 12 AM and 6 AM showed a small trend toward higher risk of MACE (IRR range: 1.11 to 1.15 per SD higher heart rate), although none of the confidence intervals exceeded one (see Figure 5.4 D).2 Across all hours, a plausible association between SDNN and heart failure was observed. However, this association disappeared after adjustment for physical acceleration and heart rate (see Figure 5.4 B).2 In contrast, heart rate between 10 PM and 9 AM was associated with heart failure (IRR range: 1.37 to 1.58 per SD higher heart rate) (see Figure 5.4 E). SDNN was consistently associated with all-cause mortality across all hours, with a stronger inverse association observed between 12 PM and 1 AM (IRR range: 0.79 to 0.88 per SD higher SDNN) (see Figure 5.4 C).2 No clear trends of association were observed between heart rate and all-cause mortality.2

SDNN preadjusted for concurrent physical acceleration and heart rate, as well as mean heart rate (HR) and physical acceleration, were measured hourly from 00:00 to 24:00. The IRR for MACE, heart failure hospitalization, and all-cause mortality are shown by hourly associations of: (A–C) preadjusted SDNN, (D–F) mean HR, and (G–I) physical acceleration. Models were adjusted for age, sex, education, alcohol consumption, smoking behavior, body mass index, total cholesterol, and HbA1c. Figure adapted from Appendix Study II2

Figure 5.4: Diurnal heart rate variability and heart rate association with major adverse cardiovascular events, heart failure, and all-cause mortality risk

5.3 Study III

5.3.1 Descriptive

In study III, 176 participants had measures of NT-proBNP. CAN was present in 31% (n = 54) of participants (40% among those with valid CAN measurements (Figure 5.5 A)).3 Meanwhile, 23% (n = 40) were unable to complete the CART assessment adequately, primarily due to insufficient air pressure during the Valsalva maneuver (n = 21).3 Compared to those without CAN, the participants with CAN were more often women (41 % vs 33 %), were more sedentary (45% vs 36%), had a higher proportion with prior major CVD (41% vs 23%) and declined eGFR (< 60ml/min/1.73 m\(^2\)) (35% vs 22%), higher levels of triglyceride (median 2.05 mmol/L vs 1.95 mmol/ L), were slightly older (median 62 years vs 61 years), had longer duration of T2D (median 19 years vs 15 years), and higher use SGLT2i (65% vs 60%) but lower use of GLP-1RA (63% vs 70%). No other difference in clinical characteristic was observed (see ?tbl-can).3

?(caption)

5.3.2 CAN and indicators of heart failure

A greater proportion of individuals with CAN were observed to exhibit elevated NT-proBNP levels (>125 pg/ml) (51.9%, n = 52/78) compared to those without CAN (23.2%, n = 26/112) (Figure 5.5 E).3 The fully adjusted odds ratio (OR) for elevated NT-proBNP in individuals with CAN was 5.69 (95% CI: 1.95; 18.49) relative to those without CAN.3 Among the CARTs, the Valsalva manoeuvre was found to demonstrate the strongest association with NT-proBNP (OR 9.00, 95% CI: 2.88; 33.09; n = 51/75), followed by deep breathing (OR 3.30, 95% CI: 1.17; 9.77; n = 33/133) and orthostatic hypertension (OR 4.04, 95% CI: 1.27; 13.77; n = 24/146).3 No significant association was identified for the lying-to-standing test (OR 0.80, 95% CI: 0.32; 1.97; n = 54/108).3 After imputation of missing CART data, the OR for CAN in relation to elevated NT-proBNP declined to 2.94 (95% CI: 1.37; 6.56).3 Sensitivity analyses, which excluded participants using beta-blockers or those with a history of CVD, resulted in a smaller sample size and wider confidence intervals, though the overall association remained unchanged.3 CAN was associated with elevated NT-proBNP in individuals both without (NYHA=I; OR = 4.3, 95% CI: 1.1; 16.3) and with heart failure symptoms (NYHA≥II; OR = 16.4, 95% CI: 1.2; 222.0), though the interaction was not statistically significant (p = 0.4).3 Similar associations were observed across WATCH-DM risk groups: very-low-to-moderate (OR = 6.1, 95% CI: 1.6; 23.5) and high-to-very-high (OR = 6.3, 95% CI: 0.83; 46.9).3 Participants with CAN were found to have 1.7 (95% CI: 0.3 to 3.0) point higher WATCH-DM risk scores compared to those without CAN.3 The OR of presenting with NYHA class II or higher was 5.51 (95% CI: 1.9 to 15.97) in the group with CAN.3

A: Percentage distribution by CAN status (no CAN, CAN). B: Percentage distribution of NT-proBNP level categories stratified by individuals with and without CAN. C: Percentage distribution of WATCH-DM risk score stratified by individuals with and without CAN. C: Percentage distribution of NYHA classification stratified by individuals with and without CAN. E: Percentage of individuals with NT-proBNP > 125 pg/ml among those with and without CAN and adjusted odds ratio from Model 4. F: Effect modification of the association between CAN and NT-proBNP by symptoms defined by NYHA classification (symptoms: NYHA ≥ II vs no symptoms: NYHA = I) and risk score defined by WATCH-DM risk (very-low-to-moderate vs high-to-very-high risk). Figure from Appendix Study III3

Figure 5.5: Distribution of NT-proBNP, NYHA Class, and WATCH-DM Score by CAN Status, and association of CAN with Elevated NT-proBNP